On elementary divisors perturbation of nonnegative matrices
-
Upload
independent -
Category
Documents
-
view
0 -
download
0
Transcript of On elementary divisors perturbation of nonnegative matrices
This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies areencouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Linear Algebra and its Applications 432 (2010) 546–555
Contents lists available at ScienceDirect
Linear Algebra and its Applications
j ourna l homepage: www.e lsev ie r .com/ loca te / laa
On elementary divisors perturbation of nonnegative
matrices�
Javier Ccapa, Ricardo L. Soto∗
Dpto. Matemáticas, Universidad Católica del Norte, Casilla, 1280, Antofagasta, Chile
A R T I C L E I N F O A B S T R A C T
Article history:
Received 23 March 2009
Accepted 1 September 2009
Available online 1 October 2009
Submitted by R.A. Brualdi
AMS classification:
15A18
Keywords:
Nonnegative matrices
Elementary divisors
Spectra perturbation
An outstanding result of Guo (1997) [3] establishes that if the list
Λ = {λ1, λ2, . . . , λn} is the spectrum of an n × n nonnegative ma-
trix, where λ1 is its Perron eigenvalue and λ2 ∈ R, then for any
t � 0, the list Λt = {λ1 + t, λ2 ± t, . . . , λn} is also the spectrum of
a nonnegative matrix. In this paper we extend the result of Guo
to elementary divisors. In particular, if A is a nonnegative matrix
with spectrum Λ then, by means of two rank one perturbations,
we construct a modified matrix B, which is also nonnegative, with
spectrum Λt and we explicitly provide the Jordan canonical form
of B.
© 2009 Elsevier Inc. All rights reserved.
1. Introduction
Let A ∈ Cn×n and let
J(A) = S−1AS =
⎡⎢⎢⎢⎣
Jn1(λ1)
Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎦
be the Jordan canonical form of A (hereafter JCF of A). The ni × ni submatrices
�Supported by Fondecyt 1085125, Chile.
∗ Corresponding author. Fax: +56 55 355599.
E-mail addresses: [email protected] (J. Ccapa), [email protected] (R.L. Soto).
0024-3795/$ - see front matter © 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.laa.2009.09.001
Author's personal copy
J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555 547
Jni(λi) =
⎡⎢⎢⎢⎢⎢⎣
λi 1
λi
. . .
. . . 1
λi
⎤⎥⎥⎥⎥⎥⎦, i = 1, 2, . . . , k
are called the Jordan blocks of J(A). Then the elementary divisors of A are the polynomials (λ − λi)ni ,
that is, the characteristic polynomials of Jni(λi), i = 1, . . . , k.A matrix A = (aij)
ni,j=1 is said to have constant row sums if all its rows add up to the same constant
α, i.e.∑n
j=1 aij = α, i = 1, . . . , n. The set of all matrices with constant row sums equal to α is denoted
by CSα . It is clear that any matrix A ∈ CSα has the eigenvector e = (1, 1, . . . , 1)T corresponding to
the eigenvalue α. Denote by ek the vector with 1 in the k-th position and zeros elsewhere, by σ(A)the spectrum of A, by ma(λ) the algebraic multiplicity of λ, and by Eij the matrix with 1 in position
(i, j) and zeros elsewhere. A nonnegative matrix A (aij � 0) is called stochastic if A ∈ CS1 and doubly
stochastic if A, AT ∈ CS1. For lack of simplicity we shall call nonnegative generalized stochastic to a
nonnegative matrix A ∈ CSα and nonnegative generalized doubly stochastic to a nonnegative matrix
Awith A, AT ∈ CSα .
In [4], Johnson proved that if Λ = {λ1, λ2, . . . , λn} is the spectrum of an n × n entrywise nonneg-
ative matrix, then Λ is also the spectrum of a nonnegative matrix with constant row sums equal to its
Perron eigenvalue. The following outstanding result of Brauer [1] has shown to be a very useful tool to
construct nonnegative matrices with prescribed spectrum.
Theorem 1.1 [1]. Let A be an n × n arbitrary matrix with eigenvalues λ1, λ2, . . . , λn. Let v = (v1, v2, . . . ,vn)
T be an eigenvector of A associated with the eigenvalue λk and let q be any n-dimensional vector. Thenthe matrix A + vqT has eigenvalues λ1, λ2, . . . , λk−1, λk + vTq, λk+1, . . . , λn.
In [3], Guo proved the following notable result:
Theorem1.2 [3]. If the list of complexnumbersΛ = {λ1, λ2, . . . , λn} is the spectrumof ann × nentrywise
nonnegative matrix, where λ1 is the Perron eigenvalue and λ2 ∈ R, then for any t � 0 the list
Λt = {λ1 + t, λ2 ± t, λ3, . . . , λn}is also the spectrum of an entrywise nonnegative matrix.
In [2], the authors showed that if Λ = {λ1, λ2, . . . , λn}, with λ1 > λ2 � · · · � λn � 0, is the spec-
trum of a positive matrix A ∈ CSλ1, then Λt = {λ1 + t, λ2 ± t, λ3, . . . , λn} is also the spectrum of a
positivematrix B ∈ CSλ1+t . Alternative results to Theorem1.2were also showed in [2] for positive gen-
eralized doubly stochastic matrices with nonnegative spectrum. Under certain constraints [2, Lemmas
2.2 and 2.7], the results still hold if Λ contains some negative numbers. The arguments used to prove
the results in [2] do not permit to relax the hypotheses to nonnegativematrices. They need of a positive
initial matrix A.
In this paper we extend the result of Guo to elementary divisors. Besides, unlike [2], our ar-
guments allow us to consider nonnegative matrices. In particular, if A is an nonnegative general-
ized stochastic matrix with spectrum Λ = {λ1, λ2, . . . , λn} and elementary divisors (λ − λ1), (λ −λ2)
n2 , . . . , (λ − λk)nk ; n2 + · · · + nk = n − 1, we construct, by means of two rank one perturba-
tions, a modified matrix B, which is also nonnegative generalized stochastic, with spectrum Λt ={λ1 + t, λ2 ± t, λ3, . . . , λn} and elementary divisors (λ − λ1 − t), (λ − λ2 ∓ t), (λ − λ2)
n2−1, (λ −λ3)
n3 , . . . , (λ − λk)nk ; n2 + · · · + nk = n − 1.
We recall the following result given in [6]:
Lemma 1.1 [6]. Let q = (q1, . . . , qn)T be an arbitrary n-dimensional vector and let A ∈ CSλ1
with JCF
Author's personal copy
548 J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555
S−1AS = J(A) =
⎡⎢⎢⎢⎣
λ1
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎦ .
Letλ1 + ∑ni=1 qi /= λi, i = 2, . . . , n. Then thematrix A + eqT has JCF J(A) + (∑n
i=1 qi)E11. In particular,
if∑n
i=1 qi = 0 then A and A + eqT are similar.
The paper is organized as follows: In Section 2 we introduce some preliminary results, which will
be used in the next section. In Section 3 we consider perturbation results, which allow us to construct,
from a nonnegative A ∈ CSλ1, a modified matrix B ∈ CSλ1+t with the desired spectral properties. We
also show, in section 4, some examples to illustrate the results.
2. Preliminary results
We start this section with the following result, which will be useful to prove the main results:
Theorem 2.1. Let the n × n matrices
A =[a xT
0 Q
]and B =
[a 0T
0 Q
],
where a ∈ C, x ∈ Cn−1, 0 = (0, . . . , 0)T and Q is an (n − 1) × (n − 1) complex matrix. If a is not an
eigenvalue of Q , then the matrices A and B are similar.
Proof. Let
S =[1 0T
e In−1
], xT = [x1, . . . , xn−1] .
Then
A = B + e1 [0, x1, . . . , xn−1]
and
SAS−1 = SBS−1 + Se1 [0, x1, . . . , xn−1] S−1
= SBS−1 + e
⎡⎣−
n−1∑i=1
xi, x1, x2, . . . , xn−1
⎤⎦ .
Observe that SBS−1, SAS−1 ∈ CSa and
J(SBS−1) =[a
J(Q)
]=
⎡⎢⎢⎢⎣
a
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎦ ,
with a /= λi, i = 2, . . . , k. Let
qT =⎡⎣−
n−1∑i=1
xi, x1, . . . , xn−1
⎤⎦ .
Then qTe = 0 and from Lemma 1.1, the matrices SBS−1, SBS−1 + eqT = SAS−1 are similar. Hence, A
and B are also similar. �
The following result is an extension of Lemma 1.1 (Lemma 3.1 in [6])
Author's personal copy
J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555 549
Lemma 2.1. Let A be an n × n complex matrix with Av = λ1v, v /= 0. Let
J(A) =
⎡⎢⎢⎢⎣
Jn1(λ1)Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎦ ; n1 + · · · + nk = n,
be the JCF of A. Let q = (q1, . . . , qn)T ∈ Cn such that λ1 + qTv is a simple eigenvalue of A + vqT . Then
J(A + vqT ) =
⎡⎢⎢⎢⎢⎢⎢⎣
λ1 + qTv
Jn1−1(λ1)Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
In particular, if n1 = 1, then J(A + vqT ) = J(A) + qTvE11. If ma(λ1) = 1 and qTv = 0 then A and
A + vqT are similar.
Proof. Let S = [v|S(2)| · · · |S(n)] be an n × n nonsingular matrix such that A = SJ(A)S−1. Then
S−1(A + vqT )S = J(A) + S−1vqT S
= J(A) + e1qT S
=
⎡⎢⎢⎢⎢⎢⎢⎣
λ1 + qTv ∗ ∗ · · · ∗Jn1−1(λ1)
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
. (1)
Hence, from Theorem 2.1,
J(A + vqT ) =
⎡⎢⎢⎢⎢⎢⎢⎣
λ1 + qTv
Jn1−1(λ1)Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
.
Moreover, if n1 = 1
J(A + vqT ) =
⎡⎢⎢⎢⎢⎣
λ1 + qTv
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎦
= J(A) + qTvE11.
If qTv = 0 andma(λ1) = 1 then from (1) and Theorem 2.1, J(A + vqT ) = J(A). �
Theorem 2.2. Let Λ = {λ1, λ2, . . . , λn} be the spectrum of an n × n nonnegative matrix A,with ρ(A) =λ1 and elementary divisors
(λ − λ1)n1 , (λ − λ2)
n2 , . . . , (λ − λk)nk ; n1 + · · · + nk = n.
Let t > 0. Then, Λt = {λ1 + t, λ2, . . . , λn} is also the spectrum of an n × n nonnegative matrix B with
elementary divisors
Author's personal copy
550 J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555
(λ − λ1 − t), (λ − λ1)n1−1, (λ − λ2)
n2 , . . . , (λ − λk)nk ; n1 + · · · + nk = n.
Proof. Let
J(A) =
⎡⎢⎢⎢⎣
Jn1(λ1)Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎦
be the JCF of A. Let Av = λ1v, with v = (v1, . . . , vn)T ∈ Rn, v /= 0, v � 0. Let t > 0. There exist a real
nonnegative vector qT = (q1, q2, . . . , qn), such that t = qTv. Then, A + vqT � 0 and from Theorem 1.1,
σ(A + vqT ) = {λ1 + t, λ2, . . . , λn} = Λt .
From Lemma 2.1,
J(A + vqT ) =
⎡⎢⎢⎢⎢⎢⎢⎣
λ1 + t
Jn1−1(λ1)Jn2(λ2)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
.
Therefore, for t > 0, Λt is the spectrum of a nonnegative matrix B = A + vqT with ρ(B) = λ1 + t
and elementary divisors
(λ − λ1 − t), (λ − λ1)n1−1, (λ − λ2)
n2 , . . . , (λ − λk)nk . �
Theorem 2.3. Let Λ = {λ1, λ2, . . . , λn} be the spectrum of an n × n nonnegative generalized doubly
stochastic matrix A, ρ(A) = λ1 and elementary divisors
(λ − λ1), (λ − λ2)n2 , . . . , (λ − λk)
nk ; n2 + · · · + nk = n − 1.
Let t > 0. Then, Λt = {λ1 + t, λ2, . . . , λn} is also the spectrum of a nonnegative generalized doubly
stochastic matrix B, ρ(B) = λ1 + t and elementary divisors
(λ − λ1 − t), (λ − λ2)n2 , . . . , (λ − λk)
nk ; n2 + · · · + nk = n − 1.
Proof. Let
J(A) =
⎡⎢⎢⎢⎣
λ1
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎦
be the JCF of A. Let q =(tn, tn, . . . , t
n
)T. Then eqT � 0 and qTe = t. Hence
A + eqT � 0,
σ(A + eqT ) = {λ1 + t, λ2, . . . , λn} = Λt ,
with
(A + eqT )e = Ae + eqTe
= λ1e + (qTe)e
= λ1e + te
= (λ1 + t)e
and
Author's personal copy
J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555 551
(A + eqT )Te = ATe + qeTe
= λ1e + nq
= λ1e + te
= (λ1 + t)e.
From Lemma 1.1, J(A + eqT ) = J(A) + tE11. Therefore, there exists B = A + eqT � 0with the required
spectral properties. �
Corollary 2.1. Let Λ = {λ1, λ2, . . . , λn} be the spectrum of an n × n symmetric nonnegative generalized
doubly stochastic matrix A, ρ(A) = λ1. Let t > 0. Then, Λt = {λ1 + t, λ2, . . . , λn} is also the spectrum
of a symmetric nonnegative generalized doubly stochastic matrix B, with ρ(B) = λ1 + t.
Proof. From Theorem 2.3 it only remains to show that B = A + eqT is symmetric. For this, observe
that eqT = qeT . Then
BT = (A + eqT )T
= AT + qeT
= A + eqT
= B. �
3. Perturbation results
In this section we introduce the main results, which extend Theorem 1.2 to elementary divisors.
Theorem 3.1. Let Λ = {λ1, λ2, . . . , λn} be the spectrum of an n × n nonnegative generalized stochastic
matrix A, with λ1 > λ2 � · · · � λp real, 2� p� n, ρ(A) = λ1 and elementary divisors
(λ − λ1), (λ − λ2)n2 , . . . , (λ − λk)
nk ; n2 + · · · + nk = n − 1.
Let t > 0. Then, Λt = {λ1 + t, λ2 + t, λ3, . . . , λn} is also the spectrum of a nonnegative generalized
stochastic matrix B, ρ(B) = (λ1 + t), and elementary divisors
(λ − λ1 − t), (λ − λ2 − t), (λ − λ2)n2−1, (λ − λ3)
n3 , . . . , (λ − λk)nk ;
with n2 + · · · + nk = n − 1.
Proof. Although it is clear, from Theorem 1.2 and the result of Johnson [4], that Λt is the spectrum of
a nonnegative generalized stochastic matrix, we here include, for the sake of completeness, this part
of the proof. Let v = (v1, v2, . . . , vn)T ∈ Rn, v /= 0, and Av = λ2v. Let
A = S
⎡⎢⎢⎢⎣
λ1
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎦ S−1,
where S = [e|v|S3| . . . |Sn]. LetvM = max {v1, v2, . . . , vn} ,vm = min {v1, v2, . . . , vn} ,
withM,m ∈ {1, . . . , n}. Let B = A + eqTe + vqT
v , where
Author's personal copy
552 J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555
qe = − tvm
vM − vmeM + tvM
vM − vmem
and
qv = t
vM − vmeM − t
vM − vmem.
Then eqTe + vqT
v � 0 and B is nonnegative. Moreover since
qTev = qT
ve = 0, qTee = qT
vv = t,
then
Be =(A + eqT
e + vqTv
)e
= Ae + eqTee + vqT
ve
= Ae + te
= (λ1 + t)e
and
(A + eqTe)v = Av + e
(qTev
)= Av = λ2v.
Hence,
σ(A + eqTe) = {λ1 + t, λ2, . . . , λn} and
σ((
A + eqTe
)+ vqT
v
)= {λ1 + t, λ2 + t, λ3, . . . , λn} = Λt
with λ2 + t = λ2 + qTvv simple eigenvalue of B. Now, from Lemma 2.1
J(A + eqTe) =
⎡⎢⎢⎢⎣
λ1 + t
Jn2(λ2). . .
Jnk(λk)
⎤⎥⎥⎥⎦ .
Let P a permutation matrix such that
PJ(A + eqTe)P
−1 =
⎡⎢⎢⎢⎢⎢⎢⎣
Jn2(λ2)λ1 + t
Jn3(λ3). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
.
Then from Lemma 2.1
J((
A + eqTe
)+ vqT
v
)=
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣
λ2 + t
Jn2−1(λ2)λ1 + t
Jn3(λ3). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
and the result follows. �
Now we consider a list Λ = {λ1, λ2, . . . , λn} ⊂ C, with p real numbers, p� n, in the following
cases:
Author's personal copy
J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555 553
(a) ma(λ1) = p or
(b) ma(λ1) = 1, ma(λ2) = p − 1
}(2)
Theorem 3.2. Let Λ = {λ1, λ2, . . . , λn}, as in (2), be the spectrum of an n × n nonnegative generalized
stochastic matrix A, ρ(A) = λ1 and elementary divisors
(λ − λ1), (λ − λ2)n2 , . . . , (λ − λk)
nk ; n2 + · · · + nk = n − 1.
Let t > 0. Then, Λt = {λ1 + t, λ2 − t, λ3, . . . , λn} is also the spectrum of a nonnegative generalized
stochastic matrix B, ρ(B) = (λ1 + t), and elementary divisors
(λ − λ1 − t)), (λ − λ2 + t)), (λ − λ2)n2−1, (λ − λ3)
n3 , . . . , (λ − λk)nk ;
n2 + · · · + nk = n − 1.
Proof. Similarly to the proof of Theorem 3.1, let (B = A + eqTe + vqT
v , where
qe = tvM
vM − vmeM − tvm
vM − vmem
and
qv = − t
vM − vmeM + t
vM − vmem.
Then B is nonnegative and Be = (λ1 + t)e. Since
B = A + eqTe + vqT
v
= SJ(A)S−1 + Se1
(qTeS
)S−1 + Se2
(qTvS
)S−1
= S[J(A) + e1
(qTeS
)+ e2
(qTvS
)]S−1,
where qTeS = (t, ∗, . . . , ∗) and qT
vS = (0,−t, ∗, . . . , ∗) and
J(A) + e1
(qTeS
)+ e2
(qTvS
)
=
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣
λ1 + t ∗ ∗ ∗ · · · ∗λ2 − t ∗ ∗ · · · ∗
Jn2−1(λ2)Jn3(λ3)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦,
then
σ(B) = {λ1 + t, λ2 − t, λ3, . . . , λn} = Λt ,
with λ1 + t, λ2 − t simple eigenvalues.
From Theorem 2.1
J(A) + e1
(qTeS
)+ e2
(qTvS
)is similar to⎡
⎢⎢⎢⎢⎢⎢⎢⎢⎣
λ1 + t
λ2 − t ∗ ∗ · · · ∗Jn2−1(λ2)
Jn3(λ3). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
Author's personal copy
554 J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555
and the matrices⎡⎢⎢⎢⎢⎢⎢⎣
λ2 − t ∗ ∗ · · · ∗Jn2−1(λ2)
Jn3(λ3). . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
and ⎡⎢⎢⎢⎢⎢⎢⎣
λ2 − t
Jn2−1(λ2)Jn3(λ3)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎦
are also similar. Then
J(A) + e1
(qTeS
)+ e2
(qTvS
)is also similar to⎡
⎢⎢⎢⎢⎢⎢⎢⎢⎣
λ1 + t
λ2 − t
Jn2−1(λ2)Jn3(λ3)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
and
J(B) =
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣
λ1 + t
λ2 − t
Jn2−1(λ2)Jn3(λ3)
. . .
Jnk(λk)
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
.
Hence B has elementary divisors
(λ − λ1 − t), (λ − λ2 + t)), (λ − λ2)n2−1, (λ − λ3)
n3 , . . . , (λ − λk)nk ;
n2 + · · · + nk = n − 1. �
4. Examples
Example 4.1. In [5, Theorem 8] it was shown that the list Λ = {6, 1, 1,−4,−4} is realizable as the
spectrum of the nonnegative matrix
A =
⎡⎢⎢⎢⎢⎢⎣
0 4 0 0 2
4 0 0 0 232
0 0 4 12
32
0 4 0 12
0 0 6 0 0
⎤⎥⎥⎥⎥⎥⎦,
which has elementary divisors (λ − 6), (λ − 1)2, (λ + 4), (λ + 4). By applying Theorem 3.1 we con-
struct, for any t � 0, a nonnegative matrix B ∈ CS6+t ,
Author's personal copy
J. Ccapa, R.L. Soto / Linear Algebra and its Applications 432 (2010) 546–555 555
B =
⎡⎢⎢⎢⎢⎢⎣
t 4 0 0 2
t + 4 0 0 0 212t + 3
20 0 4 1
2t + 1
212t + 3
20 4 0 1
2t + 1
20 0 6 0 t
⎤⎥⎥⎥⎥⎥⎦,
with Jordan canonical form J(B) = diag{6 + t, 1 + t, 1,−4,−4}.Example 4.2. In [6] it was shown that
A =
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣
1 2 2 1 1 1 1
2 0 3 1 1 1 1
3 2 0 1 1 1 1
3 2 2 0 0 1 1
0 2 2 2 0 2 1
3 2 2 1 1 0 0
1 2 2 1 1 2 0
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
has elementary divisors (λ − 9), (λ + 2)2, (λ + 1 − i)2, (λ + 1 + i)2. By applying Theorem 3.2 we
construct, for t > 0, the nonnegative matrix
B =
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣
1 t + 2 2 1 1 1 1
t + 2 0 3 1 1 1 1
3 t + 2 0 1 1 1 1
3 t + 2 2 0 0 1 1
0 t + 2 2 2 0 2 1
3 t + 2 2 1 1 0 0
1 t + 2 2 1 1 2 0
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎦
with elementary divisors (λ − 9 − t), (λ + 2 + t), (λ + 2), (λ + 1 − i)2, (λ + 1 + i)2.
References
[1] A. Brauer, Limits for the characteristic roots of amatrix IV: applications to stochasticmatrices,DukeMath. J. 19 (1952) 75–91.[2] J. Ccapa, R.L. Soto, On spectra perturbation and elementary divisors of positivematrices, Electron. J. Linear Algebra 18 (2009)
462–481.[3] W. Guo, Eigenvalues of nonnegative matrices, Linear Algebra Appl. 266 (1997) 261–270.[4] C.R. Johnson, Row stochastic matrices similar to doubly stochastic matrices, Linear Multilinear Algebra 10 (1981) 113–130.[5] R.L. Soto, O. Rojo, Applications of a Brauer theorem in the nonnegative inverse eigenvalue problem, Linear Algebra Appl.
416 (2006) 844–856.[6] R.L. Soto, J. Ccapa, Nonnegativematriceswith prescribed elementary divisors, Electron. J. Linear Algebra 17 (2008) 287–303.